# Marcus Delgado, VP of Underwriting Operations at Covridge Specialty (MGA, ~80 employees) — read of Underwriting AI, May 10 2026

> 14 years in commercial lines, moved from carrier-side to MGA three years ago, currently managing a team of six underwriters across E&O and D&O books.

## How I got here

Searched "AI underwriting automation MGA" on a Sunday night after a bad week. We had a submission backlog that pushed a renewal quote out nine days and the insured went elsewhere. I've been clicking every plausible result for about a month. This one came up on page two, which usually means it's either niche or new. I clicked because the title matched my exact job title.

## What I clicked first

The hero stopped me briefly: "evaluates applications in hours, producing defensible decisions auditors and regulators won't challenge." The word "defensible" is a real word in my world. Underwriters say "defensible" constantly when talking to E&O counsel. So whoever wrote this at least Googled the vocabulary. But "auditors and regulators won't challenge" is a guarantee no tool can make, and the fact that someone just typed that without a footnote told me something about the level of rigor behind the claims.

## Where I paused

The "Who this is NOT for" section. That's unusual for a product page and I actually read the whole thing. "Anyone who wants to launch this in under 4 weeks (it's a heavier build)" -- that's when the page shifted underneath me. This isn't a SaaS tool I can demo. This is a studio pitching me a concept they'll build for $68K. The product being sold is the idea, not the software. Once I understood that, I had to reread the entire top half with different eyes.

## What I distrusted

The "Trusted by" header with nothing after it. That's not a placeholder they forgot to fill in -- that's a design choice, and it's a bad one. It reads as: we have no customers but we needed a social proof section so we kept the header.

Also this: "Regulatory Compliance. Built for Fair Lending requirements, SOX compliance, and industry regulations. Transparent decision logic for every assessment." Fair Lending applies to consumer credit. I'm in commercial lines. SOX is a public company requirement. Neither of those is the compliance framework an MGA underwriting manager is sweating over. Whoever wrote this copied a financial-services compliance checklist without knowing which part of financial services they're selling to. That's a real tell.

The "9% probability of meaningful success, by Fermi heuristics" line buried in the build estimates is either refreshingly honest or a hedge so they can't be accused of overselling. I haven't decided which.

## What would convince me

A screenshot of an actual submission queue with timestamps showing before and after. Not a mockup, a real one with the carrier name redacted. I want to see what "hours" means: is it 2 hours or 18 hours? I also want to know what happens when the AI declines something a human underwriter would have approved. Who owns that error? What's the appeals workflow? The page doesn't mention any of that, and in my world that's 40% of the implementation problem.

A conversation with one MGA that actually ran this in production, even in a pilot, would close me faster than any case study PDF.

## What I'd ask in an email reply

1. The page mentions "Fair Lending" and "SOX" as the compliance frameworks. We're a specialty MGA, so our exposure is more around state-level surplus lines regulations and carrier-specific filing requirements. Has this been configured for that context, or is the compliance layer more generic?
2. What does "defensible decision logic" actually look like in the output? Is it a PDF? An audit trail in a system of record? A field in our AMS? Because "defensible" means nothing if it's not in the format my E&O auditor accepts.
3. The $68K to production estimate -- is that the full scope including integration with an existing AMS like Applied Epic, or is that greenfield with no integration work priced in?

## Verdict: on-the-fence

The pain is real and the vocabulary is closer than most vendor pages I've seen. But the page is selling a concept, not a tool, and it took me three reads to understand that. If someone from this studio emailed me with a specific MGA example and didn't lead with the word "AI," I'd probably take a 20-minute call.

---
*Memo by skeptic persona, generated 2026-05-10. Studio breaks own self-grading loop.*